Spatiotemporal data mining for situation awareness in microblogs

Özdikiş, Özer
Detection of real-world events using messages posted in microblogs has been the motivation of numerous recent studies. In this thesis, we study spatiotemporal data mining techniques to improve situation awareness by detecting events and estimating their locations using the content in microblogs, particularly in Twitter. We present an enhancement to the clustering techniques in the literature by measuring associations between terms in tweets in a temporal context and using these associations in a vector expansion process to improve the accuracy of online tweet clustering and event detection. Moreover, we propose a method using the Dempster-Shafer theory to estimate the locations of the detected events. We utilize three basic location-related features in tweets, namely the latitude-longitude metadata in geotagged tweets, the location names mentioned in the tweet content and the location attribute in the user profile, as independent sources of evidence. We apply combination rules in the Dempster-Shafer theory to fuse them into a single model, and estimate the whereabouts of a detected event. We demonstrate the results of our experiments for event detection and location estimation using public tweets posted in Turkey. Our experiments indicate higher success rates than those obtained by the state of the art methods.
Citation Formats
Ö. Özdikiş, “Spatiotemporal data mining for situation awareness in microblogs,” Ph.D. - Doctoral Program, Middle East Technical University, 2016.